Development of a Predictive Model to Determine Appropriate Length of Profile

Area of Science:

• Military medicine
• Data science in healthcare
• Musculoskeletal injury management

Background:

• Musculoskeletal injuries are a primary cause of soldier non-readiness, exceeding 80% of cases.
• Electronic profiles (e-Profiles) document injuries and dictate limited duty duration for soldiers.
• Previous studies indicated a median e-Profile length of 30 days.

Purpose of the Study:

• To develop a machine learning model for predicting soldier e-Profile duration.
• To optimize soldier recovery time and ensure unit readiness.

Main Methods:

• Utilized a large dataset of 2.9 million electronic profile records.
• Evaluated linear regression, decision trees, and random forests (RFs) for predictive accuracy.

Main Results:

• Random forests achieved the highest Area Under the Curve (AUC) of 0.794.
• Decision trees showed a positive predictive value of 73.6% for 0-30 day profiles.
• No model demonstrated high certainty in predicting e-Profile duration.

Conclusions:

• The evaluated machine learning models did not accurately predict e-Profile duration.
• Future research will incorporate additional data to enhance model performance.